International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. XXXIV-5/W10
IMPROVED 3D CITY MODELING WITH CYBERCITY-MODELER (CC-MODELER™)
USING AERIAL-, SATELLITE IMAGERY AND LASERSCANNER DATA
Kilian Ulm
CyberCity AG, c/o Chair of Photogrammetry
ETH Zurich Hoenggerberg, CH-8093 Zurich
Switzerland, kulm@cybercity.tv
KEY WORDS: 3D city modeling, photogrammetry, satellite imagery, laserscanner data, real-time visualisation
ABSTRACT:
Semi-automated object extraction has been established as a standard procedure for the efficient generation of 3D city models. Apart
from aerial imagery, satellite imagery and laserscanner data turned out to be an additional source to create virtual city models.
CyberCity-Modeler (CC-Modeler™) as a tool for the generation and optimisation of 3D city and plant models has been used in a
range of projects with different requirements. Therefore modules have been developed and integrated to the main software package
to improve functionality. CC-Modeler™ consists of four modules: CC-Modeler for topological structuring, CC-Edit for improving
geometry , CC-Mapping for integration of wall texture and CC-Digit for integration of digitised data from maps. Concerning the
software package CC-Modeler™, some functions as geometrical regularisation of buildings, editing functions for topology
adjustment, modeling of overhanging roofs, adding attributes to objects, mapping from texture libraries, generation of corridors etc.
are explained in this paper. Additionally, the use of satellite and laserscanner data for the efficient generation of 3D city models as an
alternative or supplement to aerial images is discussed and the progress in real-time visualisation of 3D city models is presented.
1
INTRODUCTION
CyberCity-Modeler (CC-Modeler™) uses semi-automated
object extraction from stereo-models of aerial and satellite
imagery towards the generation of realistic 3D building models.
Additionally, all objects which can be represented as polyhedral
models like digital terrain models (DTM), roads, waterways,
trees etc. can be modeled. The automatic detection of main roof
types from laserscanner data is under development.
CC-Modeler™ is a commercial software product that is
marketed by the ETH Zurich spin-off company CyberCity AG.
Realistic 3D city models are used primarily in city-, facility
planning but also in location planning of telecom companies,
3D car navigation systems, scene modeling in game industry
etc. Due to variability of applications, the software is improved
ongoing with the different project specifications and is adapted
to specific needs and requirements.
This paper discusses data capturing of point clouds from a
stereo-model for further processing with CC-Modeler™. In
addition, the four modules of the CC-Modeler™ are presented
and the workflow to improve the 3D building model is pointed
out. Furthermore, functions like generation of overhanging
roofs, adding basements and attributes are explained. Results
(incl. accuracy) from different data sources like aerial and
satellite imagery and laserscanner data are discussed.
Eventually, latest progress in real-time visualisation of 3D city
models is presented.
2
DATA CAPTURING FOR 3D CITY MODELING
WITH CC-MODELER™
Photogrammetry assists in derivation of information on surface
cover (e.g. vegetation) and man-made objects (e.g. buildings,
roads, bridges etc.). CC-Modeler™ enables generation of 3D
building models from aerial images with use of a semi-
automated method. This topology generator fits planar
structures to point clouds that are measured by a human
operator. The photogrammetric measurement and structuring is
done on Analytical Plotters or on Digital Photogrammetric
Workstations (e.g. Virtuozo, Socet Set, Z/I Image-station etc.).
The operator classifies the captured points in boundary points
and interior points, based on their different functionality in one
object (Figure 1). Different codes are assigned to these points,
which help the CC-Modeler to generate the roof faces and an
accurate topology. Each roof type can be generated and there
are no constraints regarding a roof type library.
Figure 1. Point definition in CC-Modeler.
The points of the roof boundary polygon have to be measured
clockwise or counter-clockwise (in case of the interior points,
no sequence has to be kept). The interpretation of the data
depends on the operator which gives the opportunity to achieve
any level of detail. With this method, hundreds of objects (i.e.
buildings, roads, bridges etc.) can be measured by one operator
in one day.
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. XXXIV-5/W10
3
CYBERCITY-MODELER (CC-MODELER™)
Originally, CyberCity-Modeler (CC-Modeler™) was a research
project at the ETH Zurich and was later implemented as a
commercial product from the CyberCity AG (Gruen, Wang
1998). It was designed for data acquisition and structuring for
3D city model generation. Besides structuring, the integration of
real image texture should be possible and assigning attributes to
objects should be permitted.
To fulfil project constraints, the functionality of CC-Modeler™
was extended and now consists of four modules:
CC-Modeler for topological structuring and extraction of
realistic roof and terrain texture from aerial images
CC-Edit for improving the geometry of the building model
and adding attributes or basement
CC-Mapping for integration of wall texture
CC-Digit for integration of digitised data from maps.
Figure 2 shows the work- and dataflow in the CC-Modeler™.
The point clouds that are measured on an Analytical Plotter or
on a Digital Station or taken from maps with CC-Digit, are
structured to roof faces in CC-Modeler. The data is managed in
the own V3D-format and can be edited with CC-Edit. Data
export to formats of major CAD systems (e.g. DXF, DGN etc.)
and visualisation software (e.g. FLT, VT etc.) is provided.
Figure 2. Dataflow CC-Modeler™.
3.1 CC-Modeler
As a topology generator, the CC-Modeler, which is the basis
module, generates roof topology from manually measured point
clouds. This workflow can be automatic or in step mode for
detecting face mistakes that can occur due to inadequate point
measurement.
A few main editing functions in 2D or 3D view enables solving
these problems. In the 2D view, the vector data of the 3D
objects (roof polygon) are projected on the original aerial image
or orthophoto to check how the measured data fit to the ground
(Figure 3). Through this projection, the DTM and the roofs can
be textured automatically from the aerial image.
To generate vertical walls, the roof border polygon is projected
on the DTM and intersected with it.
With this first implementation of the CyberCity-Modeler
software package, it is even possible to texture building facades
with realistic images.
With this main module, the internal data-management structure
can be translated into different formats readable by AutoCAD,
Microstation, Polytrim, ArcView and Iventor (i.e. DXF, DWG,
DGN, IV, AutoLisp etc.). Converter to visualisation formats
like FLT (Open Flight) and VT (VTree) are available.
Figure 3. Graphical User Interface of CC-Modeler.
3.2 CC-Edit
Developed as a research project, the commercial
implementation of CyberCity-Modeler has to fulfil some
specific practical requirements in an efficient way, that led to
modifications and additions in the basic software package. CCEdit as one of these additions was developed as a CAD system
for 3D city models targeted to the V3D structure.
With different editing functions, the measured objects, that are
as close as possible to their existing size and shape, can be
regularised in geometry (Paragraph 3.2.1). For the realistic
modeling of the building facades, an additional function is
integrated to create walls and overhanging roofs with the use of
cadastral maps (Paragraph 3.2.2). Even basements with arbitrary
depths can be added to buildings (Paragraph 3.2.3). The V3Dstructure allows to save information about buildings as
attributes and some geometric attributes (e.g. volume, area etc.)
can be calculated automatically (Paragraph 3.2.4).
New extensions are implemented during new project work
which improve the software and make it flexible and successful.
3.2.1 Workflow to improve geometry of 3D city models
The next paragraph is devoted to the workflow of achieving best
data quality for further processing, because in measurement,
some errors can occur by the limits of interpretation. This may
lead to edge or ridge lines that are not parallel or roof faces that
are not planar.
CC-Edit uses two strategies: automatic adjustment based on
least squares (Gruen, Wang 2001) and an approach of CAD
editing. The aim of the editing process is the generation of a 3D
model that meets the following conditions:
Same height for groups of eaves points and ridge points.
Roof faces with more than 3 points should be planar.
Collinearity of eaves and ridge lines.
Right angles in main structures.
3.2.1.1 Face adjustment and height processing
A number of query functions facilitate the detection of
differences in the model geometry. It is possible to check the
height of the two end points in a selected line that can be for
example the eaves or ridge line. To improve the quality of the
building, one can change the height of a selected point manually
or set the two points to an average height or to the height of a
selected base point.
This same height processing can also be done automatically for
the whole model with an additional face adjustment. For each
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. XXXIV-5/W10
ridge and eaves line, the average height is calculated if the
difference of the z-coordinate of the points is less than a
threshold in meters (default value in the parameter settings is
0.25 meters).
With another query it is possible to check, if the faces of a
model are coplanar. The maximum of the difference between
the face and the points is displayed and the points deviate from
the face are marked with a red rectangle, which leads the
operator to the parts of the model that have to be further edited
(tolerance is parameterised).
3.2.1.2 Grid functions
To get high quality buildings that may be combined with
cadastral data, some additional improvements are required (e.g.
orthogonality of the outline).
In case of a simple geometry, this process can be accomplished
automatically for one building or a group of buildings.
For the alignment of complex structured objects, a grid can be
set as a reference system, either automatically or manually. The
grid that consists of parallel construction lines is overlaid to the
lines that connect the measured points. The concerned object
points are snapped to the appropriate grid points which adjust
the object lines automatically to the direction of the grid lines.
The snapping box size parameter can be set according to the
different accuracy requirements.
Orthogonality, collinearity and the planar face constraint are
taken in consideration by using this editing process (Figure 4).
Figure 4. Object rectification (dashed line: before, solid line:
after).
3.2.1.3 Point fitting
There are different possibilities to fit measured points to a fixed
structure. A number of selected points can be fitted to one line.
In case of a wall that is described by a curve, the points along
the wall can be fitted to a symmetrical curve. Cylindric objects
are difficult to measure in aerial images. The border polygon of
these objects can be fitted to a circle, where the number of circle
points can be set in the input parameters providing thus
different accuracy.
If it is necessary for the improvement of the model, new points
can be added to the roof polygon and later fitted with the former
mentioned methods.
3.2.1.4 Topology adjustment
Inconsistencies in topology between adjacent buildings and
mutually overlapping roofs may arise because of measurement
errors. They can even exist after the previous explained editing
on the 3D model to improve geometry. To solve this problem,
CC-Edit offers an automated and a semi-automated procedure.
In the automated mode, the system selects a reference border
line which is kept fixed and onto which the points of the other
lines are projected perpendicularly. As reference line, the
longest line is selected, because it is assumed to be the most
stable line (Figure 5). In the semi-automated mode, this
reference line is selected manually.
Figure
5. Topology adjustment. Left: original,
automatically corrected with CC-Edit.
Right:
3.2.1.5 Roof superstructures
Roof superstructures for example windows or chimneys are
measured and managed separately. Without editing work, these
structures have walls from the roof to the ground surface after
the intersection of the roof points with the DTM. With CC-Edit
it is possible, to combine the main building with his
superstructures and cut the walls of the small objects on the roof
face in the way, that after editing, in the main building no
further senseless walls exist. Thus all objects are characterised
as standalone objects.
After intersecting, the base point of a superstructure is included
in a face of the main building. While editing the roof, this
condition can be lost, but with a simple function, the concerned
point can be moved into a chosen direction until it fits to the
face again.
3.2.2 Generating walls and overhanging roofs from
cadastral maps
In general, facades are not visible in aerial images and only the
roof surfaces can be measured. Digital cadastral maps, which
show the outer walls of buildings as part of the legal definition
of real estate property, can be used to integrate realistic facades.
Utilising this information, it is possible to model overhanging
roofs and get a higher level of detail in building modeling
(Figure 6).
Figure 6. Automated facade integration. Left: Roof polygon and
facade from map, Right: Overhanging roofs as result.
This procedure is not easy to automate. Sometimes, the roof
border polygon is not concordant with the facade border line
(Figure 7). This can occur if the facade appears shifted or
rotated relative to the roof due to measurement errors or in case
of outdated maps. Additionally it has to be mentioned, that
maps may show a lot of peripheral details like stairs which
make this procedure be handled by an operator.
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. XXXIV-5/W10
Figure 7. Roof polygons (dark) and facade representation from
maps (light).
3.2.3 Basements
As a new function, CC-Edit allows to add basements to one
house or a group of houses. The depth of the basement can be
set individually. These information can be used for geometric
calculations as described in the following paragraph on
attributes.
3.2.4 Attributes
Information about the objects can be saved as attributes. With
the attributes query & edit dialog these information can be
enquired and modified. A number of objects can be combined
to a complex object for saving the same attributes.
CC-Edit allows to calculate geometric information for planning
purposes. At that moment the following geometric attributes are
implemented (other calculations can be implemented on
request):
Height (of eaves line)
Upboard area (area of eaves)
Ground lowest/highest height
Basement-lowest height
Base-map area
Volume
Figure 8. Graphical User Interface of CC-Mapping.
3.3.2 Corridors
This function uses the fact, that a corridor looks most realistic,
when it is created exactly according to the facade image, where
the walkthrough is visible. In CC-Mapping the model including
the facade images is loaded and displayed. If the wall, where the
corridor should be created, is selected, the image appears on the
right window of the software and the border of the walkthrough
can be digitised and is generated automatically by the software
(Figure 9). Thus, arcades can be generated with minimum
effort.
Solutions for the management of 3D city models in a spatial
information system are proposed (Gruen, Wang 1999; Wang,
Gruen 2000). An implementation of this idea in a commercial
GIS is in progress.
3.3 CC-Mapping
The module CC-Mapping was developed for fast texturing of
the walls and roofs of the 3D city models in our format V3D.
Recently, a tool has been developed, that uses the textured
image to create corridors through a building part.
3.3.1 Wall/Roof Texturing
The roofs can be automatically textured with realistic texture
from the aerial images with CC-Modeler or can be mapped with
colour in CC-Mapping, as same for the walls (Figure 8).
Most realistic textured models can be get through wall pictures
that are taken on site with a digital camera. Obstacles like trees,
cars or people and the perspective view of the camera makes
image processing necessary. After mapping an image, a link to
the corresponding image is created in the V3D file, comparable
to the VRML format.
For generic texturing, a library of texture images can be used.
With minimum effort, the walls of a model can be textured with
sample images, that best fits to the environment of the model,
when realistic textures are not required. The sample images
represent only a part of a specific wall type. For texturing, this
part has to be repeated correctly on the wall.
The implementation of a random distribution of generic textures
to the whole model is in progress.
Figure 9. Corridors. Left: Original textured model, Center:
Generated corridor, Right: Corridor in wireframe
mode.
3.4 CC-Digit
For planning purposes, it is important to be able to create 3D
models of future buildings on the basis of maps and
architectural schemes because they cannot be measured in the
actual aerial image.
With CC-Digit, the border polygon of the planned buildings can
be captured easily by digitising the border points. This module
creates the same file format like the data capturing system with
an analytical plotter or a digital station which can be handled by
CC-Modeler and be converted to different formats like VRML
or the internal format V3D.
This module provides the fast generation of buildings that have
to be visualised in a planning process.
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. XXXIV-5/W10
4
DATA SOURCE
Aerial, satellite imagery and laserscanner data are used as raw
data (see paragraph 2) whereas depending on requirements,
different approaches are applied.
4.1 Aerial imagery
At that moment, aerial images are the most common used raw
data (Figure 10). For capturing the 3D point cloud for further
processing in CC-Modeler™, the stereo pairs of the images are
needed. The scale of the images depends on the accuracy that is
required for the 3D model and is normally about 1:5000 with a
forward and a side overlap of 30 and 60 percent respectively. If
the images are used for True-Orthophoto, the side overlap is
suggested to be 60 percent.
Using this data, many building details can be measured from the
aerial images and the measurement error is maximal 0,2 meter
in height.
4.3 Laserscanner data
For the generation of 3D city models from laserscanner data, a
density of laserscanner points of more than 2 points/sqm are
required. Big areas are already surveyed with laserscanner,
which is seen as an advantage for the application of this data.
The procedure for the calculation of 3D building models from
laserscanner data uses a tangential plane as a first
approximation that suits the laserscanner points. From this
geometric model, edge lines are derived whereas edge lines of
building structures are generated (e.g. eaves lines, ridge lines
etc.).These lines are combined to faces to represent the roof
planes. It is possible to automatically derive standard roof types
like saddleback roofs or pavilion roofs (Steidler et al. 2002).
This part is still under development (see example in Figure 12).
The accuracy is expected to be 0.3-0.5 meters in height.
Figure 12. 3D city model (derived from laserscanner data).
5
Figure 10. 3D city model Karlsruhe (derived from aerial
imagery). Courtesy of Vermessungsamt Karlsruhe.
4.2 Satellite imagery
In case of large areas, recently high resolution satellite imagery
is used, like the 1-meter panchromatic from Ikonos (Figure 11).
The data capturing process is the same as with aerial images,
but the accuracy is less, measurement error can be up to 1 meter
in height. DTM and Orthophoto can be derived automatically.
REAL-TIME VISUALISATION
Visualisation is performed by real-time tools like TerrainView
(www.viewtec.ch) or CyberWalk (www.muellersystec.de).
TerrainView and CyberWalk are two competitive products with
interactive user interfaces. In particular, they are able to display
terrain and buildings as separate objects. Free navigation within
a selected geographic area are allowed. Foreground and
background are displayed with different data subsets which
means that a level-of-detail (LOD) is provided and performance
and storage capacity is optimised. Data is compressed and
decompressed during visualisation. Features like flight paths,
weather simulation etc. can be provided.
For high quality visualisation, the use of a True-Orthophoto is
recommended to avoid that the roof texture is visible on the
ground beside the building.
In the meantime, even the real-time visualisation of the interior
of buildings is possible (Figure 13).
Figure 11. 3D city model Izmir (derived from satellite imagery
Ikonos).
Figure 13. Interior of a church visualised with CyberWalk.
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. XXXIV-5/W10
6
CONCLUSIONS
CC-Modeler™ is a tool that uses a semi-automated object
extraction from photogrammetric stereo models to generate
realistic 3D building models. It is independent from the
photogrammetric data capturing system and with the module
CC-Edit a powerful tool to modify and improve the geometry of
3D models is provided.
The possibility of the integration of cadastral maps for the
generation of overhanging roofs, the mapping options and the
option to generate corridors from facade images makes 3D city
models more realistic.
Hence CC-Modeler™ is a unique tool for generating 3D city
models efficiently and with a high degree of detail and accuracy
and the option of attributation means a next step to a threedimensional GIS.
7
REFERENCES
Gruen, A., Wang, X. 1998. CC-Modeler: a topology generator
for 3-D city models, ISPRS Journal of Photogrammetry &
Remote Sensing 53(5): pp. 286-295
Gruen, A., Wang, X. 1999. CyberCity Spatial Information
System (SIS): A new concept for the management of 3D city
models in a hybrid GIS. Proc. 20th Asian Conference on Remote
Sensing, Hongkong, November 1999, pp. 121-128
Wang, X., Gruen, A. 2000. A hybrid GIS for 3-D city models.
International Archives of Photogrammetry and Remote Sensing,
vol. 33, part 4/3, pp. 1165-1172
Gruen, A., Wang, X. 2001. News from CyberCity-Modeler.
Invited Paper, Third International Workshop on Automatic
Extraction of Man-Made Objects from Aerial and Space
Images, Monte Verita, Ascona, Switzerland, 10-15 June 2001
Gruen, A., Steidler, F, Wang, X. 2002. Generation and
visualization of 3D-city and facility models using CyberCity
Modeler. MapAsia 2002, Bankok, August 2002
Steidler F., Ulm K., Wild E., Müller K. 2002. Ableitung von
3D-Gebäudemodellen aus Photogrammetrie-, Satelliten- und
Laserscannerdaten, interaktive Begehung in Echtzeit.
Pressemitteilung InterGeo 2002, Frankfurt, Oktober 2002